Association does not establish causation.

 

In clinical medicine most approaches to level of evidence are related to changing outcomes so in this context highest-level evidence for  Causation = reliable evidence that an intervention or exposure increases or decreases the likelihood of an outcome.  High-quality evidence or level 1 evidence (or whatever the “top” of a level of evidence rating system uses) means there is a high likelihood of a causal relationship.

 

A study design that provides evidence of association (but no information on the temporal relationship between the “exposure” and the “outcome”) is not reliable because the two variables could both be “outcomes” related to an unrecognized exposure (confounding variable) and there is a high risk of bias in concluding a causal or direct relationship between the two variables.

 

But a study showing an association is some evidence suggesting greater likelihood for causation than not having an association.  If this is the best evidence we have for the concept it is useful to know that this is the best evidence we have and how much/how little one can rely on this evidence.

 

The value and clinical use of this information will vary with the context. 

·         An association of exposure to a new drug and a resolution of an unrelated clinical problem – perhaps a suggestion of efficacy – we would typically not suggest clinical use in most cases (but we might if the drug is considered safe and the clinical problem is considered serious with no alternative solutions).  In most cases we would want additional evidence before suggesting clinical use.

·         An association of exposure to a new drug and a previously unrecognized seriously adverse outcome – if serious enough with a strong enough association we may move to cautions or even contraindications without waiting for better evidence of serious harm.

 

 

So as others have said it is useful to know the finding is an “association” and not overinterpret it as causation, but also know there is “some evidence” and not simply conclude “no evidence” because it doesn’t reach the highest-quality type of evidence.

 

 

To translate that into a level of evidence (the original question) it depends on what level of evidence system you are using.  There are many different approaches and some systems would have a place in the hierarchy of evidence below the “bottom” absence of evidence assignment.

 

 

Of course it gets a little more complicated when you start to apply this to actual data.  If analyzing a “cross-sectional study” is there some evidence of temporal relationships because (a) cross-sectional was used as a descriptor for how participants were enrolled, but data gathering went beyond the timeframe of enrollment (so it is actually a cohort study or case-control study) or (b) certain factors are recognized as having temporal expectations regardless of time of data gathering (a genetic variable would generally be an “exposure” and not an “outcome” and we assume a person’s genetic findings today represent their genetic findings at birth so do not need to get “old data” for such a variable)

 

 

Brian S. Alper, MD, MSPH, FAAFP

Founder of DynaMed
Vice President of EBM Research and Development, Quality & Standards

dynamed.ebscohost.com